Technical thread for Server Discussion (merged)

[der_kluge]

I'd like to echo what Spectre72 said. I'm a data architect, and they pay me an obscene amount of money to analyze databases, design databases, etc.

If possible, I'd like to analyze the table layouts and the database schema that we are using for MySQL. There might be ways to increase performance through better database design, which might not necessarily have to warrant any code changes.

 

[Psionicist]

Hey Psionicist!

What are the axes of that diagram, again?

Thank you,
-George

The Y-axis is number of IO-operations per second, higher is better. The X-axis is "load", or number of outstanding asynchornous I/Os per disk, think of it as number of threads doing disk-access simultaneously. 64 is heavy load for a server, 256 is _very heavy_ load (such as disk-swapping).

 

[Arnix]

Has anyone considered cacheing (something like Squid) the non-dynamic pages? I know the forums need to stay dynamic, but the front page (and several other high traffic pages) don't change often enough to warrant being completely dynamic.

Cacheing those pages will/should have a significant positive impact on the number of DB hits. I know that once we implemented this at work, the load on our Oracle DB dropped significantly (once the cached pages got loaded).

Btw, squid is open source and can be used for enterprise level cacheing.

- Cameron

 

[Michael Morris]

Has anyone considered cacheing (something like Squid) the non-dynamic pages? I know the forums need to stay dynamic, but the front page (and several other high traffic pages) don't change often enough to warrant being completely dynamic.

Cacheing those pages will/should have a significant positive impact on the number of DB hits. I know that once we implemented this at work, the load on our Oracle DB dropped significantly (once the cached pages got loaded).

Btw, squid is open source and can be used for enterprise level cacheing.

- Cameron

The news page is driven by the same software that drives the forums. If you look closely you'll note that it presents *your* user information, current server time and so on. Add to that the multiple style sheets and...

Maintaining a news site without a database backend is a pain in the ass. If you don't believe me, ask Eric Noah. The original pages had no db backend.

 

[Michael Morris]

I'd like to echo what Spectre72 said. I'm a data architect, and they pay me an obscene amount of money to analyze databases, design databases, etc.

If possible, I'd like to analyze the table layouts and the database schema that we are using for MySQL. There might be ways to increase performance through better database design, which might not necessarily have to warrant any code changes.

The vbulletin database has 4 years development behind it and is used on sites much larger than ours. I doubt the database structure is the problem.

 

[Michael Morris]

This can be solved by an accelerator, such as turck mmcache or PHP Accelerator.

Definately intersted in those, as this site is highly PHP intensive.

I forget who asked, but the OS we are using and intend to continue using is Linux, though I'm unsure of the exact flavor (and wouldn't reveal it even if I knew).

 

[Enkhidu]

In my experience, dual CPUs are the way to go (assuming that you don't end up doing the blade server thing to make expansion easier down the road). I've seen a dual 1.4 P3 box run right by a single 2.8 P4 (both running SCSI, too).

 

[Michael Morris]

In my experience, dual CPUs are the way to go (assuming that you don't end up doing the blade server thing to make expansion easier down the road). I've seen a dual 1.4 P3 box run right by a single 2.8 P4 (both running SCSI, too).

Sorry, but I disagree. Dual CPU's in one box aren't effective as two distinct boxes because they have to share resources (memory, hard drive). Also each box can be optimized to it's given task be it database, PHP or image serving.

 

[BSF]

I think Enkhidu was voicing an opinion that a dual-processor box can outperform a single-processor box, even when the single processor is at least twice as fast as either processor in the dual-processor box. Depending on how the OS and the applications behave, this is quite true. Though some applications will not derive any additional benefit from an additional processor.

 

[Michael Morris]

I think Enkhidu was voicing an opinion that a dual-processor box can outperform a single-processor box, even when the single processor is at least twice as fast as either processor in the dual-processor box. Depending on how the OS and the applications behave, this is quite true. Though some applications will not derive any additional benefit from an additional processor.

Yes, that is true - but most dual processor boxes cost almost as much as two seperate boxes. If you're willing to spend the $4000 on a dual processor box you might as well spend $4200 on two machines.